Crate cognis_llm 

cognis-llm

v2-beta LLM client. Re-exports error, schemars, and Message from cognis-core so v2 user code can target a single crate_path = "cognis_llm" and have all macro-generated paths resolve.

Re-exports

pub use schema::schema_for_tool;

pub use chat::ChatOptions;

pub use chat::ChatResponse;

pub use chat::HealthStatus;

pub use chat::StreamChunk;

pub use chat::ToolCallDelta;

pub use chat::Usage;

pub use tools::BaseTool;

pub use tools::SchemaBasedTool;

pub use tools::Tool;

pub use tools::ToolDefinition;

pub use tools::ToolInput;

pub use tools::ToolOutput;

pub use tools::ToolRegistry;

pub use client::Client;

pub use client::ClientBuilder;

pub use factory::ProviderConstructor;

pub use factory::ProviderRegistry;

pub use factory::ProviderSpec;

pub use provider::openrouter::OpenRouterBuilder;

pub use provider::openrouter::OpenRouterProvider;

pub use provider::wrappers::Capability;

pub use provider::wrappers::ChatInterceptor;

pub use provider::wrappers::CircuitBreakerProvider;

pub use provider::wrappers::CircuitState;

pub use provider::wrappers::CircuitStats;

pub use provider::wrappers::FailureClassifier;

pub use provider::wrappers::FnChatInterceptor;

pub use provider::wrappers::GracefulDegradationProvider;

pub use provider::wrappers::InterceptorProvider;

pub use provider::wrappers::LoadBalancerProvider;

pub use provider::wrappers::LoadBalancingStrategy;

pub use provider::wrappers::ProviderRoute;

pub use provider::wrappers::RandomStrategy;

pub use provider::wrappers::RetryableClassifier;

pub use provider::wrappers::RoundRobinStrategy;

pub use provider::wrappers::RoutingProvider;

pub use provider::wrappers::RoutingStrategy;

pub use provider::wrappers::WeightedRoundRobinStrategy;

pub use provider::LLMProvider;

pub use provider::Provider;

pub use streaming::Aggregated;

pub use streaming::StreamAggregator;

pub use usage::UsageTracker;

pub use structured::StructuredClient;

pub use cognis_core::schemars;

Modules

chat

Chat-completion request/response types shared across providers.

client

User-facing Client. Holds an Arc<dyn LLMProvider> and dispatches through it. Implements Runnable<Vec<Message>, Message> so it composes inside graphs.

error

Error module re-exported from cognis-core.

factory

Provider factory + runtime registry.

message

Message types for LLM conversations.

prelude

Common imports for v2 user code building agents and tools.

provider

Provider enum + LLMProvider trait. Closed enum, not an open registry — adding a provider means editing the enum.

schema

Schema generation tuned for OpenAI / Anthropic / Ollama tool calling.

streaming

Stream-chunk aggregation utilities.

structured

Structured output: turn a Client into a Runnable<Vec<Message>, T> for any T: JsonSchema + DeserializeOwned.

tools

Tool trait + ergonomic tiers + supporting types.

usage

Running-total usage tracker for budget enforcement.

Macros

schema_for

Generates a RootSchema for the given type using default settings.

simple_tool

Build an Arc<dyn Tool> inline.

Structs

AiMessage

An AI/assistant message, optionally carrying tool call requests.

HumanMessage

A human/user message.

SystemMessage

A system prompt or instruction message.

ToolCall

One tool invocation requested by the LLM in an AiMessage.

ToolMessage

A tool execution result message.

Enums

Message

A single message in an LLM conversation.

Traits

JsonSchema

A type which can be described as a JSON Schema document.

Derive Macros

JsonSchema